Bayesian models and the parametric G-formula in time-dependent settings: robustness and sensitivity analysis

• Activity Number: 296 - Bayesian Biostatistical Applications
• Type: Contributed
• Date/Time: Tuesday, August 1, 2017 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Arielle K Anglin* and Jason A Roy
• Companies: Department of Biostatistics, University of Pennsylvania and University of Pennsylvania
• Keywords: Bayesian ; causal inference ; g-formula ; parametric models ; unmeasured confounding ; observational study

Abstract

The parametric g-formula was developed to estimate average causal effects in studies with time-dependent confounding. While several recent papers have illustrated the use of the g-formula in practice, these have almost exclusively been from a Frequentist perspective. In this paper we focus on a Bayesian approach for modeling observational data. We carry out simulation studies to assess the robustness of estimated treatment effects under various violations of parametric modeling assumptions in a time dependent setting, and compare the performance with semiparametric approaches. We also develop a sensitivity analysis for the no unmeasured confounding assumption in the Bayesian setting. We apply the methods to data on the comparative effectiveness of treatments for inflammatory bowel disease.